How to Install ComfyUI on Windows 11 (Step-by-Step Guide)

ComfyUI runs well on Windows and the installation process is straightforward once you know what you’re doing. This guide walks through everything from prerequisites to your first generated image, including installing ComfyUI Manager so you can add custom nodes without faffing around with Git commands manually.

Prerequisites

Before you start, you need three things installed on your system:

• Python 3.10, 3.11, or 3.12 — Python 3.13 is not currently recommended as some dependencies lag behind. Download from python.org and during installation, tick Add Python to PATH.
• Git — Download from git-scm.com. Accept the defaults during installation.
• NVIDIA GPU drivers — If you have an NVIDIA card, make sure your drivers are reasonably up to date. Check via Device Manager or the NVIDIA app.

Open a new Command Prompt or PowerShell window after installing Python and Git, and verify both are on your PATH:

python --version
git --version

Both should return version numbers. If they don’t, restart your terminal or check your PATH environment variable.

Cloning the Repository

Navigate to the folder where you want to install ComfyUI. Somewhere with plenty of space — models alone can run to 10–20GB or more. Then clone the repo:

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

Installing Python Dependencies

ComfyUI uses a standard requirements.txt. Install the dependencies with pip:

pip install -r requirements.txt

If you have an NVIDIA GPU, you should also install PyTorch with CUDA support. The command varies depending on your CUDA version, but for CUDA 12.1 (the most common current version):

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Check your CUDA version via nvidia-smi in a terminal. If you see CUDA 11.8, substitute cu118 in the URL above. Installing the wrong version won’t break anything — it’ll just fall back to CPU — but you want CUDA for GPU acceleration.

Downloading a Model

ComfyUI without a model won’t generate anything. Models go in the models/checkpoints/ folder inside your ComfyUI directory. For a first test, Stable Diffusion 1.5 is small and fast:

• Go to huggingface.co and search for runwayml/stable-diffusion-v1-5
• Download v1-5-pruned-emaonly.ckpt (about 4GB)
• Place it in ComfyUI/models/checkpoints/

For SDXL or Flux models, see the separate guides linked at the end of this post — they require additional supporting files.

Launching ComfyUI

From inside the ComfyUI folder:

python main.py

ComfyUI will start and print a local URL to the terminal — typically http://127.0.0.1:8188. Open this in your browser. You should see the node-based interface with a default workflow already loaded.

If you want it to open automatically in your browser, add the --auto-launch flag:

python main.py --auto-launch

Common Launch Flags

A few flags worth knowing:

• --lowvram — Enables aggressive VRAM optimisation. Use if you have 6GB or less.
• --cpu — Forces CPU-only mode. Very slow, but useful for testing on a machine without a GPU.
• --listen 0.0.0.0 — Makes ComfyUI accessible on your local network, not just localhost. Useful for accessing it from another device.
• --port 8189 — Changes the port if 8188 is in use.

Installing ComfyUI Manager

ComfyUI Manager is the community plugin manager for ComfyUI. It adds an in-browser interface for installing, updating, and removing custom nodes without touching the command line. This is almost essential — install it before you do anything else.

Stop ComfyUI if it’s running. Navigate into the custom_nodes folder:

cd custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git
cd ..
python main.py

Once ComfyUI restarts, you’ll see a Manager button in the top-right corner of the interface. Click it to access the full custom node library.

Your First Image

With a model loaded and ComfyUI running, the default workflow is already wired up. Click on the Load Checkpoint node and select your downloaded model from the dropdown. Type a prompt into the positive CLIP Text Encode node, then click Queue Prompt. Your first image will appear in the output node on the right.

If you get an error about the model not being found, double-check that the .ckpt or .safetensors file is in models/checkpoints/ and restart ComfyUI.

Keeping ComfyUI Updated

ComfyUI is actively developed. To pull the latest changes:

git pull
pip install -r requirements.txt

Run this from the ComfyUI root folder periodically. ComfyUI Manager also has an update option for the manager itself and any installed custom nodes.

Troubleshooting Common Issues

• CUDA out of memory: Add --lowvram or --medvram to the launch command. For very tight VRAM, also try --medvram-sdxl.
• pip install fails: Make sure you’re using Python 3.10–3.12, not 3.13. Check with python --version.
• No GPU detected: Run python -c "import torch; print(torch.cuda.is_available())". If it prints False, reinstall PyTorch with the correct CUDA version.
• Blank output or black images: Usually a model compatibility issue or a broken VAE. Try a different model or add a VAE node explicitly.

Next Steps

Once you’re up and running, the next steps are installing a better model and learning the node interface. For Flux.1 installation, see our guide at How to Install Flux in ComfyUI. For a walkthrough of the interface itself, see the ComfyUI Beginner’s Guide.

Running ComfyUI on CPU (No GPU)

If your machine does not have a dedicated GPU, or you have an AMD card on Windows, you can still run ComfyUI on CPU. Add the --cpu flag when launching: python main.py --cpu. CPU generation is significantly slower — a single 512×512 image that takes 5 seconds on a mid-range GPU will take several minutes on CPU — but it works for testing and learning the interface without needing NVIDIA hardware. AMD GPU users on Windows should also look into DirectML support: launch with --directml instead of the default CUDA path. Performance will be lower than CUDA but substantially better than pure CPU.

Where to Store Models

By default, ComfyUI looks for checkpoints in the models/checkpoints/ folder inside your ComfyUI directory. VAEs go in models/vae/, LoRAs in models/loras/, and ControlNet models in models/controlnet/. If you already have models downloaded for Automatic1111, you can point ComfyUI at the same folder rather than duplicating files. Copy the file extra_model_paths.yaml.example in the ComfyUI root to extra_model_paths.yaml and edit it to include your A1111 models folder path.

For a full index of every ComfyUI guide on Serverman, see the ComfyUI complete guide and hub.